The state of AI companion apps, in 23 datapoints.

The AI companion category exited 2023 as an unregulated frontier. By September 2026 it has a real-money market, real users, and, for the first time, real law. Industry trackers put the global AI companion market somewhere between $37B and $49B in size as of 2025–2026, with Grand View Research estimating roughly $36.8B in 2025 and Fortune Business Insights projecting $49.5B by 2026. Precedence Research separately projects the market compounding at roughly 33.8% annually through 2030. Estimates vary widely depending on how analysts define "companion" (the high and low ends differ by an order of magnitude), so we report the range rather than a single number.

Usage data is more consistent. A Common Sense Media study in 2025 found that 72% of U.S. teens had tried an AI companion at least once, with 52% using them regularly. Character.AI, the largest companion platform by a wide margin (though mostly non-NSFW), reports a median user spending roughly 90 minutes per day in-app. The audience for the category we cover in our dataset (NSFW-adjacent, paid) is narrower but shares the same demographic shape: overwhelmingly 18–34, mobile-dominant, and heavily concentrated in the US, UK, and Germany.

On the regulatory side, 2026 is the year the rules arrived. California's SB 243, signed October 2025 and effective January 1, 2026, was the first state law in the US to specifically regulate AI companion chatbots. It requires clear disclosure that the user is talking to an AI, a reminder every three hours for users the app knows are minors, crisis-resource protocols for self-harm content, and steps to keep sexually explicit images away from minors. Several other states passed their own chatbot laws during 2026, and since August 2026 the EU AI Act also requires chatbots to disclose that they are AI. At the federal level, the bipartisan GUARD Act (introduced October 2025) would go further and keep minors off AI companions entirely. It is not law yet. The practical effect on our dataset: every app with a US-visible sign-up is now adding friction to the onboarding flow, and several have quietly retuned their filters since the start of the year.

Across the 21 paid apps in the dataset, entry-tier monthly pricing spans a 10× range, from $5.95 at the cheapest (SpicyChat.ai) to $30.00/month at the most expensive (Grok (companions retired)). The median tier sits at $12.99/month, and the mean is pulled above it to $14.23 by a small cluster of premium tiers ($20+).

Annual pricing is where the real discounts live. 4 of 12 apps with publicly listed annual plans are priced at least 50% below their monthly rate on an effective basis. The most aggressive: MyLovely AI and GoLove AI at roughly $4.15/month on annual (a discount of well over 60% against their standard monthly rates), and Candy.ai at $3.99/month effective (70% off its $13.99 monthly). If you intend to keep using the category for more than four months, monthly billing is effectively a tax on optionality.

Lifetime pricing is the other unusual wrinkle. 2 apps in the dataset offer a one-time lifetime purchase: Kupid.ai ($777), Get-Harder ($99.99). Get-Harder's $99.99 lifetime is the pricing outlier: it pays for itself in ~7 months of monthly use. Kupid's $777 sits at the opposite end, pitched at heavy users who'd otherwise spend $350+/year.

Every app in the dataset ships a free tier. That's now table stakes, not a differentiator. Image generation is close to ubiquitous: 23 of 24 apps offer some form of it, and 13 score 8+ on image quality. The interesting axes are the ones that still separate apps.

Voice is the scarcest premium feature. Only 10 apps (42%) rate 7+ on voice, and only two (Muah.ai on ULTRA VIP and Nectar on premium tiers) ship genuinely real-time calls with competitive latency. The rest of the category is still stuck on pre-recorded voice messages. DarLink.AI is the one app that outsources its voice stack (it uses ElevenLabs), and the quality difference is audible.

Video generation is the 2024–2026 breakthrough. 14 apps (58%) now generate short NSFW video clips inline with chat, a feature that essentially did not exist commercially at the start of 2024. Secrets AI, MyLovely AI, Xtease.AI, DarLink AI, and Lovescape are the clearest entrants. Video quality is still uneven (short clips, occasional motion artefacts), but the gap between "does not exist" and "ships as a headline feature" closed inside 18 months.

End-to-end encryption claims remain rare (1 of 24) and unverified. OurDream AI markets E2E as a first-class feature, and Secrets AI's homepage now claims it too. Neither has a published audit. Discreet billing, where the app avoids showing up as an adult merchant on your card statement, is similarly scarce: 2 apps explicitly offer it. DarLink AI goes the other way and warns users its descriptor is explicit.

The launch-year shape tells a clear story. 13 of the 24 apps in the dataset shipped in 2023, and that cohort still holds the top of the leaderboard (Candy.ai, DreamGF.ai, Muah.ai, DarLink AI, OurDream AI). 10 launched in 2024, and they cluster slightly lower on overall ratings but are the ones pushing the newer features: video generation (Secrets AI, MyLovely AI, Lovescape), novel UX (Swipey.AI's swipe-based discovery, though Swipey has since pivoted toward a creator/fans platform model), and multi-LLM architectures (Joi AI).

What the chart doesn't show: the launch rate has outpaced our ability to test new entrants, and we curate heavily. This dataset excludes apps that folded within six months, apps with no meaningful user base, and a dozen near-clones with almost identical feature sets. The true number of "AI girlfriend / companion" apps that have launched since January 2023 is several hundred; the number that are still operating and worth paying for is closer to our sample size.

Muah is no longer the only app in the dataset with a breach on HaveIBeenPwned. MyLovely AI was added on April 8, 2026: 106,300 accounts, with email addresses and Discord/X usernames, and reporting says user prompts and links to generated images were exposed. HIBP flags it as sensitive. Separately, 404 Media reported in November 2025 that misconfigured cloud storage at Secret Desires exposed nearly 2 million images and videos, including face-swap content made from photos of real women. The other 21 apps have no known public incidents as of September 2026, which is not the same as "are definitely secure." It's the same as "have not been publicly caught." That distinction matters for a category where the data in question is, almost by definition, data users would prefer not to be tied back to their real identity.

The privacy-score distribution above reflects our rubric weighting: data-retention policy, breach history, signup friction, and corporate transparency. The category's norm is 6–7, and a confirmed incident drops a provider a clear band below that: MyLovely AI scores a 3, Muah and Secret Desires a 5. The apps that score 8+ are those with either no third-party data handling (Janitor's BYO-key model), explicit E2E claims (OurDream), or conservative data practices and clean public track records.

The rating distribution is bimodal-adjacent: 1 app clears 9.0, 9 sit in the strong 8.0–8.9 band, 13 occupy the solid 7.0–7.9 middle, and 1 rate below 7.0. There's no long tail of 2s and 3s. That's a curation artefact, not a market reality. Many launched apps we tested never made the dataset because they fell below our publish threshold.

Price does not predict score. The average monthly price for apps rated 8.0+ is $16.29, and for apps rated below 8.0 it's $12.35. That's essentially noise at 24 data points. In this category, what you pay for is features and polish, not the underlying chat quality. Our cheapest paid pick (SpicyChat.ai, $5.95/mo) scores inside half a point of apps costing three times as much.